The workflow output_prompt is passed off to a 2nd agent via oneshot; that agent has no conversational context and is really just doing a style check. It is pretty much blind to the actual workflow and the context. This makes the output_prompt a bit misleading to the naked eye. For example: let's say in my output_prompt I indicate that citations must be included in the output. If those instructions are not included at the workflow stage, the agent formatting the result via the output_prompt will not have the citations in the first place!.
The workflow
output_promptis passed off to a 2nd agent viaoneshot; that agent has no conversational context and is really just doing a style check. It is pretty much blind to the actual workflow and the context. This makes theoutput_prompta bit misleading to the naked eye. For example: let's say in myoutput_promptI indicate that citations must be included in the output. If those instructions are not included at the workflowstage, the agent formatting theresultvia theoutput_promptwill not have the citations in the first place!.